Descripción de la oferta
Merista builds the physical AI workforce for food production. This is our founding team robotics hire, reporting to the CTO, owning the robot's mind, and everything between intent and motion. Full-time and on-site in Barcelona. The Challenge Food changes shape, sticks to tools, and behaves differently as containers empty. A successful movement can still produce an incorrect portion. A policy that works in a demonstration may struggle after hours of operation. That means the hard problems are the whole job: How do we connect perception and manipulation when appearance alone doesn't tell us how food will behave? How do we control portion weight while meeting cycle-time requirements? How do we detect a failed transfer and recover without stopping the line? How do we collect demonstrations and production data that lead to measurable improvements? We expect robot learning to play an important role. You'll help determine where it earns its place, how to evaluate it, and how it should work with sensing, control, and mechanical design. What you'd own You'll lead technical choices across manipulation, learning, and control, working with perception and mechatronics engineers to define the interfaces. You'll own the path from a policy's output to reliable motion on the robot, including real-time execution, failure detection, and recovery. Your responsibility continues into deployment, where we find out which assumptions survive. Execution and feedback control: trajectory generation, contact-rich manipulation, force control, and portion-weight feedback. You'll connect policy outputs to real-time control, manage inference latency and timing, and define how the robot responds when an action cannot execute as intended. Failure detection and recovery: detecting failed transfers and interrupted actions, bringing the robot to a safe state, and resuming work reliably. You'll build automatic recovery where practical and give a remote operator the context and control needed to intervene. Manipulation policies and learning: deciding what the policy outputs, how it is trained and evaluated, and where imitation learning, reinforcement learning, simulation, or conventional control earn their place with limited fleet data. Data and evaluation: capturing useful demonstrations, failures, and recoveries, and building repeatable tests that show whether changes improve accuracy, speed, and reliability. What you bring Hands-on experience building and running control software on real robot arms, including trajectory execution and feedback from sensors. You have taken manipulation from a demonstration into repeated, reliable operation. Practical understanding of control-loop timing, latency, and fault handling. You can explain a failure you diagnosed on a physical robot, how you made it recover, and how you verified the fix. Practical understanding of robot learning or related ML, including data quality, evaluation, and deployment. Strong Python and enough systems understanding to debug across sensors, models, control, and hardware. Evidence that you can take an ambiguous problem, design useful experiments, and learn from failures. Interest in working closely with mechanical and mechatronics engineers. You don't need prior food-industry experience. Show us something you built, what broke, and how you improved it. About us Most meals in the developed world pass through a food factory. The hardest and most repetitive manipulation tasks on the line still defeat machines, like portioning and assembling real food in small batches that frequently change. So people do them by hand in cold and wet rooms. Factories cannot keep those rooms staffed. Merista builds the physical AI workforce for food production: operated robots that take over these tasks. One robot rolls onto a factory floor without rebuilding the line. We deploy it and run it, and the factory pays for the work it does. We've already built and operated a multi-site robotic fleet for restaurants, dealt with production failures, field operations, remote recovery, hardware iteration, deployments and customers. Now we're starting again with a much more ambitious technical thesis. Our long-term vision is the self-driving food factory. The work today is one task at a time. Why join at this stage The core technical choices are still open. Your work will shape the data we collect, how we measure progress, and what we deploy. Our longer-term thesis is that useful factory deployments can create a compounding learning advantage: observe real failures, capture the right data, improve the system, and carry those improvements into the next deployment. Building that loop is part of this role. As the team grows, you can deepen your technical ownership or move toward team leadership. We value both paths. Expect hardware delays, repeated experiments, and production debugging alongside new development. Part of the job is standing on a factory floor at 6am, watching a robot fail, and working with the team to understand what needs to change. We always find ways to adapt, improvise, and overcome. If you enjoy owning the whole path from an idea to a working machine, there is a great deal here to build.
